© 2011 Cisco Systems, Inc. All rights reserved. APRICOT 2011 1 BGP Techniques for Internet Service Providers Philip Smith <[email protected]> APRICOT 2011 Hong Kong, SAR, China 15 - 25 February 2011
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 1
BGP Techniques for Internet ServiceProviders
Philip Smith <[email protected]>APRICOT 2011Hong Kong, SAR, China15 - 25 February 2011
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 2
Presentation Slides
Will be available onftp://ftp-eng.cisco.com/pfs/seminars/APRICOT2011-BGP-Techniques.pdfAnd on the APRICOT 2011 website
Feel free to ask questions any time
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 3
BGP Techniques for Internet ServiceProviders
BGP Basics
Scaling BGP
Using Communities
Deploying BGP in an ISP network
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 5
Border Gateway Protocol
A Routing Protocol used to exchange routinginformation between different networks
Exterior gateway protocol
Described in RFC4271RFC4276 gives an implementation report on BGPRFC4277 describes operational experiences using BGP
The Autonomous System is the cornerstone of BGPIt is used to uniquely identify networks with a common routingpolicy
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 6
Autonomous System (AS)
Collection of networks with same routing policy
Single routing protocol
Usually under single ownership, trust and administrative control
Identified by a unique 32-bit integer (ASN)
AS 100
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 7
Autonomous System Number (ASN)
Two ranges0-65535 (original 16-bit range)65536-4294967295 (32-bit range - RFC4893)
Usage:0 and 65535 (reserved)1-64495 (public Internet)64496-64511 (documentation - RFC5398)64512-65534 (private use only)23456 (represent 32-bit range in 16-bit world)65536-65551 (documentation - RFC5398)65552-4294967295 (public Internet)
32-bit range representation specified in RFC5396Defines “asplain” (traditional format) as standard notation
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 8
Autonomous System Number (ASN)
ASNs are distributed by the Regional InternetRegistries
They are also available from upstream ISPs who are membersof one of the RIRs
Current 16-bit ASN allocations up to 58367 have beenmade to the RIRs
Around 3600 are visible on the Internet
Each RIR has also received a block of 32-bit ASNsOut of 1063 assignments, around 600 are visible on the Internet
See www.iana.org/assignments/as-numbers
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 9
AS 100 AS 101
AS 102
EE
BB DD
AA CC
Peering
BGP Basics
Runs over TCP – port 179
Path vector protocol
Incremental updates
“Internal” & “External” BGP
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 10
AS 100 AS 101
AS 102
DMZNetwork
AA
BB
CC
DD
EE
Shared network between ASes
Demarcation Zone (DMZ)
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 11
BGP General Operation
Learns multiple paths via internal and external BGPspeakers
Picks the best path and installs in the forwarding table
Best path is sent to external BGP neighbours
Policies are applied by influencing the best pathselection
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 12
eBGP & iBGP
BGP used internally (iBGP) and externally (eBGP)
iBGP used to carrySome/all Internet prefixes across ISP backboneISP’s customer prefixes
eBGP used toExchange prefixes with other ASesImplement routing policy
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 13
BGP/IGP model used in ISP networks
Model representation
IGP
iBGP
IGP
iBGP
IGP
iBGP
IGP
iBGP
eBGP eBGP eBGP
AS1 AS2 AS3 AS4
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 14
External BGP Peering (eBGP)
Between BGP speakers in different AS
Should be directly connected
Never run an IGP between eBGP peers
AS 100 AS 101CC
AA
BB
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 15
Internal BGP (iBGP)
BGP peer within the same AS
Not required to be directly connectedIGP takes care of inter-BGP speaker connectivity
iBGP speakers must to be fully meshed:They originate connected networksThey pass on prefixes learned from outside the ASNThey do not pass on prefixes learned from other iBGPspeakers
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 16
Internal BGP Peering (iBGP)
Topology independent
Each iBGP speaker must peer with every other iBGPspeaker in the AS
AS 100
AA
DD
CC
BB
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 17
Peering to Loopback Interfaces
Peer with loop-back interfaceLoop-back interface does not go down – ever!
Do not want iBGP session to depend on state of a single interfaceor the physical topology
AS 100
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 19
AS-Path
Sequence of ASes aroute has traversed
Used for:Loop detectionApplying policy
AS 100
AS 300
AS 200
AS 500
AS 400
170.10.0.0/16 180.10.0.0/16
150.10.0.0/16
180.10.0.0/16 300 200 100170.10.0.0/16 300 200150.10.0.0/16 300 400
180.10.0.0/16 300 200 100170.10.0.0/16 300 200
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 20
AS-Path (with 16 and 32-bit ASNs)
Internet with 16-bit and32-bit ASNs
32-bit ASNs are 65536and above
AS-PATH lengthmaintained
180.10.0.0/16 300 23456 23456170.10.0.0/16 300 23456
AS 70000
AS 300
AS 80000
AS 90000
AS 400
170.10.0.0/16 180.10.0.0/16
150.10.0.0/16
180.10.0.0/16 300 80000 70000170.10.0.0/16 300 80000150.10.0.0/16 300 400
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 21
AS 100
AS 300
AS 200
AS 500
170.10.0.0/16 180.10.0.0/16
180.10.0.0/16 300 200 100170.10.0.0/16 300 200140.10.0.0/16 300
140.10.0.0/16 500 300170.10.0.0/16 500 300 200
140.10.0.0/16
AS-Path loop detection
180.10.0.0/16 is notaccepted by AS100 as theprefix has AS100 in its AS-PATH – this is loopdetection in action
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 22
160.10.0.0/16
150.10.0.0/16
150.10.1.1 150.10.1.2
AS 100
AS 300AS 200
AA BB
CC
150.10.0.0/16 150.10.1.1160.10.0.0/16 150.10.1.1
eBGP
iBGP
Next Hop
eBGP – address of external neighbour iBGP – NEXT_HOP from eBGP Mandatory non-transitive attribute
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 23
AS 300
BBCC
120.1.1.0/24 120.1.254.2120.1.2.0/23 120.1.254.3
iBGP120.1.1.0/24
120.1.2.0/23
Loopback120.1.254.2/32
Loopback120.1.254.3/32
AA
DD
iBGP Next Hop
Next hop is ibgp router loopback address Recursive route look-up
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 24
120.68.1.0/24
150.1.1.3
150.1.1.1
150.1.1.2
120.68.1.0/24 150.1.1.3
AS 201
AS 200
CC
AA BB
Third Party Next Hop
eBGP between Router Aand Router C
eBGP between RouterA andRouterB
120.68.1/24 prefix has nexthop address of 150.1.1.3 –this is passed on to RouterCinstead of 150.1.1.2
More efficient No extra config needed
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 25
Next Hop Best Practice
BGP default is for external next-hop to be propagatedunchanged to iBGP peers
This means that IGP has to carry external next-hopsForgetting means external network is invisibleWith many eBGP peers, it is unnecessary extra load on IGP
ISP Best Practice is to change external next-hop to bethat of the local router
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 26
Next Hop (Summary)
IGP should carry route to next hops
Recursive route look-up
Unlinks BGP from actual physical topology
Change external next hops to that of local router
Allows IGP to make intelligent forwarding decision
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 27
Origin
Conveys the origin of the prefix
Historical attributeUsed in transition from EGP to BGP
Transitive and Mandatory Attribute
Influences best path selection
Three values: IGP, EGP, incompleteIGP – generated by BGP network statementEGP – generated by EGPincomplete – redistributed from another routing protocol
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 28
Aggregator
Conveys the IP address of the router or BGP speakergenerating the aggregate route
Optional & transitive attribute
Useful for debugging purposes
Does not influence best path selection
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 29
Local Preference
AS 400
AS 200
160.10.0.0/16AS 100
AS 300
160.10.0.0/16 500> 160.10.0.0/16 800
500 800 EE
BB
CC
AA
DD
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 30
Local Preference
Non-transitive and optional attribute
Local to an AS – non-transitiveDefault local preference is 100 (Cisco IOS)
Used to influence BGP path selectiondetermines best path for outbound traffic
Path with highest local preference wins
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 31
Multi-Exit Discriminator (MED)
AS 201
AS 200
120.68.1.0/24
AA BB120.68.1.0/24 1000120.68.1.0/24 2000
CC DD
120.68.1.0/24 2000> 120.68.1.0/24 1000
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 32
Multi-Exit Discriminator
Inter-AS – non-transitive & optional attribute
Used to convey the relative preference of entry pointsdetermines best path for inbound traffic
Comparable if paths are from same ASImplementations have a knob to allow comparisons of MEDsfrom different ASes
Path with lowest MED wins
Absence of MED attribute implies MED value of zero(RFC4271)
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 33
Multi-Exit Discriminator“metric confusion”
MED is non-transitive and optional attributeSome implementations send learned MEDs to iBGP peers bydefault, others do notSome implementations send MEDs to eBGP peers by default,others do not
Default metric varies according to vendorimplementation
Original BGP spec (RFC1771) made no recommendationSome implementations handled absence of metric as meaninga metric of 0Other implementations handled the absence of metric asmeaning a metric of 232-1 (highest possible) or 232-2Potential for “metric confusion”
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 34
Community
Communities are described in RFC1997Transitive and Optional Attribute
32 bit integerRepresented as two 16 bit integers (RFC1998)Common format is <local-ASN>:xx0:0 to 0:65535 and 65535:0 to 65535:65535 are reserved
Used to group destinationsEach destination could be member of multiple communities
Very useful in applying policies within and betweenASes
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 35
Community Example(before)
permit 160.10.0.0/16 out
ISP 1permit 100.10.0.0/16 in
XX
ISP 2
100.10.0.0/16
AS 300
AS 400FF
EE
permit 170.10.0.0/16 out
AS 200
permit 170.10.0.0/16 in
BB
170.10.0.0/16
permit 160.10.0.0/16 in
AS 100 AA
160.10.0.0/16
CC
DD
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 36
Community Example(after)
160.10.0.0/16 300:1
ISP 1100.10.0.0/16 300:9
XX
ISP 2
100.10.0.0/16
AS 300
AS 400FF
EE
170.10.0.0/16 300:1
AS 200
170.10.0.0/16 300:1
BB
170.10.0.0/16
160.10.0.0/16 300:1
AS 100 AA
160.10.0.0/16
CC
DD
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 37
Well-Known Communities
Several well known communitieswww.iana.org/assignments/bgp-well-known-communities
no-export 65535:65281do not advertise to any eBGP peers
no-advertise 65535:65282do not advertise to any BGP peer
no-export-subconfed 65535:65283do not advertise outside local AS (only used withconfederations)
no-peer 65535:65284do not advertise to bi-lateral peers (RFC3765)
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 38
105.7.0.0/16105.7.X.X No-Export
105.7.0.0/16
AS 100 AS 200
105.7.X.X
CC FF
GG
DDAA
BB EE
No-Export Community
AS100 announces aggregate and subprefixesIntention is to improve loadsharing by leaking subprefixes
Subprefixes marked with no-export community Router G in AS200 does not announce prefixes with no-export
community set
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 39
No-Peer Community
Sub-prefixes marked with no-peer community are not sent to bi-lateralpeers
They are only sent to upstream providers
105.7.0.0/16105.7.X.X No-Peer
105.7.0.0/16
AA
BB
EE
DD
CC
C&D&E arepeers e.g.
Tier-1s
upstream
upstream
upstream
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 40
What about 4-byte ASNs?
Communities are widely used for encoding ISP routingpolicy
32 bit attribute
RFC1998 format is now “standard” practiceASN:number
Fine for 2-byte ASNs, but 4-byte ASNs cannot beencoded
Solutions:Use “private ASN” for the first 16 bitsWait for www.ietf.org/internet-drafts/draft-ietf-idr-as4octet-extcomm-generic-subtype-02.txt to be implemented
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 41
CommunityImplementation details
Community is an optional attributeSome implementations send communities to iBGP peers bydefault, some do notSome implementations send communities to eBGP peers bydefault, some do not
Being careless can lead to community “confusion”ISPs need consistent community policy within their own networksAnd they need to inform peers, upstreams and customers abouttheir community expectations
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 42
BGP Path Selection Algorithm
Why Is This the Best Path?
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 43
BGP Path Selection Algorithm for IOSPart One
Do not consider path if no route to next hop
Do not consider iBGP path if not synchronised (CiscoIOS only)
Highest weight (local to router)
Highest local preference (global within AS)
Prefer locally originated route
Shortest AS path
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 44
BGP Path Selection Algorithm for IOSPart Two
Lowest origin codeIGP < EGP < incomplete
Lowest Multi-Exit Discriminator (MED)If bgp deterministic-med, order the paths before comparing
(BGP spec does not specify in which order the paths shouldbe compared. This means best path depends on order inwhich the paths are compared.)
If bgp always-compare-med, then compare for all pathsotherwise MED only considered if paths are from the same AS(default)
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 45
BGP Path Selection Algorithm for IOSPart Three
Prefer eBGP path over iBGP path
Path with lowest IGP metric to next-hop
Lowest router-id (originator-id for reflected routes)
Shortest Cluster-ListClient must be aware of Route Reflector attributes!
Lowest neighbour IP address
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 46
BGP Path Selection Algorithm
In multi-vendor environments:Make sure the path selection processes are understood foreach brand of equipmentEach vendor has slightly different implementations, extra steps,extra features, etcWatch out for possible MED confusion
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 47
Applying Policy with BGP
Controlling Traffic Flow & Traffic Engineering
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 48
Applying Policy in BGP:Why?
Network operators rarely “plug in routers and go”
External relationships:Control who they peer withControl who they give transit toControl who they get transit from
Traffic flow control:Efficiently use the scarce infrastructure resources (external linkload balancing)Congestion avoidanceTerminology: Traffic Engineering
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 49
Applying Policy in BGP:How?
Policies are applied by:Setting BGP attributes (local-pref, MED, AS-PATH, community),thereby influencing the path selection processAdvertising or Filtering prefixesAdvertising or Filtering prefixes according to ASN and AS-PATHsAdvertising or Filtering prefixes according to Communitymembership
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 50
Applying Policy with BGP:Tools
Most implementations have tools to apply policies toBGP:
Prefix manipulation/filteringAS-PATH manipulation/filteringCommunity Attribute setting and matching
Implementations also have policy language which cando various match/set constructs on the attributes ofchosen BGP routes
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 52
BGP Capabilities
Documented in RFC2842
Capabilities parameters passed in BGP open message
Unknown or unsupported capabilities will result inNOTIFICATION message
Codes:0 to 63 are assigned by IANA by IETF consensus64 to 127 are assigned by IANA “first come first served”128 to 255 are vendor specific
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 53
BGP Capabilities
Current capabilities are: 0 Reserved [RFC3392]
1 Multiprotocol Extensions for BGP-4 [RFC4760]
2 Route Refresh Capability for BGP-4 [RFC2918]
3 Outbound Route Filtering Capability [RFC5291]
4 Multiple routes to a destination capability [RFC3107]
5 Extended Next Hop Encoding [RFC5549]
64 Graceful Restart Capability [RFC4724]
65 Support for 4 octet ASNs [RFC4893]
66 Deprecated
67 Support for Dynamic Capability [ID]
68 Multisession BGP [ID]
69 Add Path Capability [ID]
See www.iana.org/assignments/capability-codes
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 54
BGP Capabilities
Multiprotocol extensionsThis is a whole different world, allowing BGP to support morethan IPv4 unicast routesExamples include: v4 multicast, IPv6, v6 multicast, VPNsAnother tutorial (or many!)
Route refresh is a well known scaling technique –covered shortly
32-bit ASNs have recently arrived
The other capabilities are still in development or notwidely implemented or deployed yet
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 55
BGP for Internet Service Providers
BGP Basics
Scaling BGP
Using Communities
Deploying BGP in an ISP network
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 57
BGP Scaling Techniques
Original BGP specification and implementation was finefor the Internet of the early 1990s
But didn’t scale
Issues as the Internet grew included:Scaling the iBGP mesh beyond a few peers?Implement new policy without causing flaps and route churning?Keep the network stable, scalable, as well as simple?
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 58
BGP Scaling Techniques
Current Best Practice Scaling TechniquesRoute RefreshPeer-groupsRoute Reflectors (and Confederations)
Deploying 4-byte ASNs
Deprecated Scaling TechniquesRoute Flap Damping
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 59
Dynamic Reconfiguration
Route Refresh
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 60
Route Refresh
BGP peer reset required after every policy changeBecause the router does not store prefixes which are rejectedby policy
Hard BGP peer reset:Terminates BGP peering & Consumes CPUSeverely disrupts connectivity for all networks
Soft BGP peer reset (or Route Refresh):BGP peering remains activeImpacts only those prefixes affected by policy change
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 61
Route Refresh Capability
Facilitates non-disruptive policy changes
For most implementations, no configuration is neededAutomatically negotiated at peer establishment
No additional memory is used
Requires peering routers to support “route refreshcapability” – RFC2918
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 62
Consider the impact to beequivalent to a router reboot
Dynamic Reconfiguration
Use Route Refresh capability if supportedfind out from the BGP neighbour status displayNon-disruptive, “Good For the Internet”
If not supported, see if implementation has aworkaround
Only hard-reset a BGP peering as a last resort
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 63
Route Reflectors
Scaling the iBGP mesh
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 64
Two solutionsRoute reflector – simpler to deploy and runConfederation – more complex, has corner case advantages
Avoid ½n(n-1) iBGP mesh
Scaling iBGP mesh
n=1000 ⇒ nearlyhalf a million
ibgp sessions!
14 routers = 91iBGP sessions
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 65
AS 100
Route Reflector: Principle
AA
CCBB
Route Reflector
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 66
AS 100
AA
BB CC
Clients
Reflectors
Route Reflector
Reflector receives pathfrom clients and non-clients
Selects best path
If best path is from client,reflect to other clients andnon-clients
If best path is fromnon-client, reflect to clientsonly
Non-meshed clients
Described in RFC4456
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 67
Route Reflector: Topology
Divide the backbone into multiple clusters
At least one route reflector and few clients per cluster
Route reflectors are fully meshed
Clients in a cluster could be fully meshed
Single IGP to carry next hop and local routes
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 68
Route Reflector: Loop Avoidance
Originator_ID attributeCarries the RID of the originator of the route in the local AS(created by the RR)
Cluster_list attributeThe local cluster-id is added when the update is sent by the RRBest to set cluster-id is from router-id (address of loopback)(Some ISPs use their own cluster-id assignment strategy – butneeds to be well documented!)
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 69
Route Reflector: Redundancy
Multiple RRs can be configured in the same cluster –not advised!
All RRs in the cluster must have the same cluster-id (otherwiseit is a different cluster)
A router may be a client of RRs in different clustersCommon today in ISP networks to overlay two clusters –redundancy achieved that way→ Each client has two RRs = redundancy
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 70
Route Reflectors:Redundancy
AS 100
Cluster One
Cluster Two
PoP2PoP1
PoP3
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 71
Route Reflector: Benefits
Solves iBGP mesh problem
Packet forwarding is not affected
Normal BGP speakers co-exist
Multiple reflectors for redundancy
Easy migration
Multiple levels of route reflectors
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 72
Route Reflector: Deployment
Where to place the route reflectors?Always follow the physical topology!This will guarantee that the packet forwarding won’t be affected
Typical ISP network:PoP has two core routersCore routers are RR for the PoPTwo overlaid clusters
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 73
Route Reflector: Migration
Typical ISP network:Core routers have fully meshed iBGPCreate further hierarchy if core mesh too big
Split backbone into regions
Configure one cluster pair at a timeEliminate redundant iBGP sessionsPlace maximum one RR per clusterEasy migration, multiple levels
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 74
AS 200
AS 100
AS 300AA
BB
GGFFEE
DD
CC
Route Reflector: Migration
Migrate small parts of the network, one part at a time
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 76
Confederations
Divide the AS into sub-ASeBGP between sub-AS, but some iBGP information is kept
Preserve NEXT_HOP across thesub-AS (IGP carries this information)Preserve LOCAL_PREF and MED
Usually a single IGP
Described in RFC5065
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 77
Confederations (Cont.)
Visible to outside world as single AS – “ConfederationIdentifier”
Each sub-AS uses a number from the private AS range (64512-65534)
iBGP speakers in each sub-AS are fully meshedThe total number of neighbours is reduced by limiting the fullmesh requirement to only the peers in the sub-ASCan also use Route-Reflector within sub-AS
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 78
Confederations
Configuration (Router C):router bgp 65532 bgp confederation identifier 200 bgp confederation peers 65530 65531 neighbor 141.153.12.1 remote-as 65530 neighbor 141.153.17.2 remote-as 65531
AS 200
Sub-AS65530
Sub-AS65532 Sub-AS
65531C B
A
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 79
Confederations: AS-Sequence
Sub-ASSub-AS6500265002
Sub-ASSub-AS6500365003
Sub-ASSub-AS6500165001
Confederation100
GG
Sub-ASSub-AS6500465004
CC
DD EE
BB
180.10.0.0/16 200
180.10.0.0/16 {65002} 200
AA
180.10.0.0/16 {65004 65002} 200
HH FF
180.10.0.0/16 100 200
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 80
Route Propagation Decisions
Same as with “normal” BGP:From peer in same sub-AS → only to external peers
From external peers → to all neighbors
“External peers” refers toPeers outside the confederationPeers in a different sub-AS
Preserve LOCAL_PREF, MED and NEXT_HOP
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 81
InternetConnectivity
Multi-LevelHierarchy
Policy Control Scalability
MigrationComplexity
Confederations
RouteReflectors
Anywherein the
NetworkYes Yes
Yes
RRs or Confederations
YesAnywhere
in theNetwork
Medium
Very High Very Low
Mediumto High
Most new service provider networks now deploy Route Reflectors from Day One
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 82
More points about Confederations
Can ease “absorbing” other ISPs into you ISP – e.g., ifone ISP buys another
Or can use AS masquerading feature available in someimplementations to do a similar thing
Can use route-reflectors with confederation sub-AS toreduce the sub-AS iBGP mesh
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 83
Deploying 32-bit ASNs
How to support customers using the extended ASN range
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 84
32-bit ASNs
Standards documentsDescription of 32-bit ASNs
www.rfc-editor.org/rfc/rfc4893.txtTextual representation
www.rfc-editor.org/rfc/rfc5396.txtNew extended community
www.rfc-editor.org/rfc/rfc5668.txt
AS 23456 is reserved as interface between 16-bit and32-bit ASN world
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 85
32-bit ASNs – terminology
16-bit ASNsRefers to the range 0 to 65535
32-bit ASNsRefers to the range 65536 to 4294967295(or the extended range)
32-bit ASN poolRefers to the range 0 to 4294967295
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 86
Getting a 32-bit ASN
Sample RIR policywww.apnic.net/docs/policy/asn-policy.html
From 1st January 200732-bit ASNs were available on request
From 1st January 200932-bit ASNs were assigned by default16-bit ASNs were only available on request
From 1st January 2010No distinction – ASNs assigned from the 32-bit pool
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 87
Representation
Representation of 0-4294967295 ASN rangeMost operators favour traditional format (asplain)A few prefer dot notation (X.Y):
asdot for 65536-4294967295, e.g 2.4asdot+ for 0-4294967295, e.g 0.64513
But regular expressions will have to be completely rewritten forasdot and asdot+ !!!
For example:^[0-9]+$ matches any ASN (16-bit and asplain)This and equivalents extensively used in BGP multihomingconfigurations for traffic engineering
Equivalent regexp for asdot is: ^([0-9]+)|([0-9]+\.[0-9]+)$
Equivalent regexp for asdot+ is: ^[0-9]+\.[0-9]+$
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 88
Changes
32-bit ASNs are backward compatible with 16-bit ASNs There is no flag day You do NOT need to:
Throw out your old routersReplace your 16-bit ASN with a 32-bit ASN
You do need to be aware that:Your customers will come with 32-bit ASNsASN 23456 is not a bogon!You will need a router supporting 32-bit ASNs to use a 32-bitASN locally
If you have a proper BGP implementation, 32-bit ASNswill be transported silently across your network
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 89
How does it work?
If local router and remote router supports configurationof 32-bit ASNs
BGP peering is configured as normal using the 32-bit ASN
If local router and remote router does not supportconfiguration of 32-bit ASNs
BGP peering can only use a 16-bit ASN
If local router only supports 16-bit ASN and remoterouter/network has a 32-bit ASN
Compatibility mode is initiated…
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 90
Compatibility Mode:
Local router only supports 16-bit ASN and remote router uses 32-bit ASN
BGP peering initiated:Remote asks local if 32-bit supported (BGP capability negotiation)When local says “no”, remote then presents AS23456Local needs to be configured to peer with remote using AS23456
BGP peering initiated (cont):BGP session established using AS2345632-bit ASN included in a new BGP attribute called AS4_PATH
(as opposed to AS_PATH for 16-bit ASNs)
Result:16-bit ASN world sees 16-bit ASNs and 23456 standing in for 32-bitASNs32-bit ASN world sees 16 and 32-bit ASNs
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 91
180.10.0.0/16 123 23456 23456170.10.0.0/16 123 23456
AS 80000
AS 123
AS 70000
AS 90000
AS 321
170.10.0.0/16 180.10.0.0/16
150.10.0.0/16
180.10.0.0/16 123 70000 80000170.10.0.0/16 123 70000150.10.0.0/16 123 321
Example:
Internet with 32-bit and16-bit ASNs
AS-PATH lengthmaintained
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 92
What has changed?
Two new BGP attributes:AS4_PATH
Carries 32-bit ASN path infoAS4_AGGREGATOR
Carries 32-bit ASN aggregator infoWell-behaved BGP implementations will simply pass thesealong if they don’t understand them
AS23456 (AS_TRANS)
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 93
asdotformat
asplainformat
What do they look like?
IPv4 prefix originated by AS196613as4-7200#sh ip bgp 145.125.0.0/20BGP routing table entry for 145.125.0.0/20, version 58734Paths: (1 available, best #1, table default) 131072 12654 196613 204.69.200.25 from 204.69.200.25 (204.69.200.25) Origin IGP, localpref 100, valid, internal, best
IPv4 prefix originated by AS3.5as4-7200#sh ip bgp 145.125.0.0/20BGP routing table entry for 145.125.0.0/20, version 58734Paths: (1 available, best #1, table default) 2.0 12654 3.5 204.69.200.25 from 204.69.200.25 (204.69.200.25) Origin IGP, localpref 100, valid, internal, best
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 94
TransitionAS
What do they look like?
IPv4 prefix originated by AS196613But 16-bit AS world view:
BGP-view1>sh ip bgp 145.125.0.0/20
BGP routing table entry for 145.125.0.0/20, version 113382
Paths: (1 available, best #1, table Default-IP-Routing-Table)
23456 12654 23456
204.69.200.25 from 204.69.200.25 (204.69.200.25)
Origin IGP, localpref 100, valid, external, best
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 95
If 32-bit ASN not supported:
Inability to distinguish between peer ASes using 32-bit ASNsThey will all be represented by AS23456Could be problematic for transit provider’s policy
Inability to distinguish prefix’s origin ASHow to tell whether origin is real or fake?The real and fake both represented by AS23456(There should be a better solution here!)
Incorrect NetFlow summaries:Prefixes from 32-bit ASNs will all be summarised under AS23456Traffic statistics need to be measured per prefix and aggregatedMakes it hard to determine peerability of a neighbouring network
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 96
Implementations (Feb 2011)
Cisco IOS-XR 3.4 onwards
Cisco IOS-XE 2.3 onwards
Cisco IOS 12.0(32)S12, 12.4(24)T, 12.2SRE, 12.2(33)SXI1 onwards
Cisco NX-OS 4.0(1) onwards
Quagga 0.99.10 (patches for 0.99.6)
OpenBGPd 4.2 (patches for 3.9 & 4.0)
Juniper JunOSe 4.1.0 & JunOS 9.1 onwards
Redback SEOS
Force10 FTOS7.7.1 onwards
http://as4.cluepon.net/index.php/Software_Support for a complete list
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 97
Route Flap Damping
Network Stability for the 1990s
Network Instability for the 21st Century!
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 98
Route Flap Damping
For many years, Route Flap Damping was a stronglyrecommended practice
Now it is strongly discouraged as it appears to causefar greater network instability than it cures
But first, the theory…
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 99
Route Flap Damping
Route flapGoing up and down of path or change in attribute
BGP WITHDRAW followed by UPDATE = 1 flapeBGP neighbour going down/up is NOT a flap
Ripples through the entire InternetWastes CPU
Damping aims to reduce scope of route flappropagation
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 100
Route Flap Damping (continued)
RequirementsFast convergence for normal route changesHistory predicts future behaviourSuppress oscillating routesAdvertise stable routes
Implementation described in RFC 2439
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 101
Operation
Add penalty (1000) for each flapChange in attribute gets penalty of 500
Exponentially decay penaltyhalf life determines decay rate
Penalty above suppress-limitdo not advertise route to BGP peers
Penalty decayed below reuse-limitre-advertise route to BGP peerspenalty reset to zero when it is half of reuse-limit
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 102
Operation
Reuse limit
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
0
1000
2000
3000
4000
Time
Penalty
Suppress limit
NetworkAnnounced
NetworkRe-announced
NetworkNot Announced
Penalty
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 103
Operation
Only applied to inbound announcements from eBGPpeers
Alternate paths still usable
Controllable by at least:Half-lifereuse-limitsuppress-limitmaximum suppress time
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 104
Configuration
Implementations allow various policy control with flapdamping
Fixed damping, same rate applied to all prefixesVariable damping, different rates applied to different ranges ofprefixes and prefix lengths
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 105
Route Flap Damping History
First implementations on the Internet by 1995
Vendor defaults too severeRIPE Routing Working Group recommendations in ripe-178,ripe-210, and ripe-229http://www.ripe.net/ripe/docsBut many ISPs simply switched on the vendors’ default valueswithout thinking
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 106
Serious Problems:
"Route Flap Damping Exacerbates Internet RoutingConvergence“
Zhuoqing Morley Mao, Ramesh Govindan, George Varghese &Randy H. Katz, August 2002
“What is the sound of one route flapping?”Tim Griffin, June 2002
Various work on routing convergence by Craig Labovitzand Abha Ahuja a few years ago
“Happy Packets”Closely related work by Randy Bush et al
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 107
Problem 1:
One path flaps:BGP speakers pick next best path, announce to all peers, flapcounter incrementedThose peers see change in best path, flap counter incrementedAfter a few hops, peers see multiple changes simply caused bya single flap → prefix is suppressed
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 108
Problem 2:
Different BGP implementations have different transittime for prefixes
Some hold onto prefix for some time before advertisingOthers advertise immediately
Race to the finish line causes appearance of flapping,caused by a simple announcement or path change →prefix is suppressed
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 109
Solution:
Do NOT use Route Flap Damping whatever you do!
RFD will unnecessarily impair accessto your network andto the Internet
More information contained in RIPE Routing WorkingGroup recommendations:
www.ripe.net/ripe/docs/ripe-378.[pdf,html,txt]
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 110
BGP for Internet Service Providers
BGP Basics
Scaling BGP
Using Communities
Deploying BGP in an ISP network
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 111
Service Provider use of Communities
Some examples of how ISPs make life easier forthemselves
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 112
BGP Communities
Another ISP “scaling technique”
Prefixes are grouped into different “classes” orcommunities within the ISP network
Each community means a different thing, has a differentresult in the ISP network
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 113
BGP Communities
Communities are generally set at the edge of the ISPnetwork
Customer edge: customer prefixes belong to differentcommunities depending on the services they have purchasedInternet edge: transit provider prefixes belong to differencecommunities, depending on the loadsharing or trafficengineering requirements of the local ISP, or what the demandsfrom its BGP customers might be
Two simple examples follow to explain the concept
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 114
Community Example: Customer Edge
This demonstrates how communities might be used atthe customer edge of an ISP network
ISP has three connections to the Internet:IXP connection, for local peersPrivate peering with a competing ISP in the regionTransit provider, who provides visibility to the entire Internet
Customers have the option of purchasing combinationsof the above connections
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 115
Community Example: Customer Edge
Community assignments:IXP connection: community 100:2100Private peer: community 100:2200
Customer who buys local connectivity (via IXP) is put in community100:2100
Customer who buys peer connectivity is put in community100:2200
Customer who wants both IXP and peer connectivity is put in100:2100 and 100:2200
Customer who wants “the Internet” has no community setWe are going to announce his prefix everywhere
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 116
CORE
Aggregation Router
CustomersCustomersCustomers
Border Router
Community Example: Customer Edge
Communities set at the aggregation router where the prefix isinjected into the ISP’s iBGP
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 117
Community Example: Customer Edge
No need to alter filters at the network border whenadding a new customer
New customer simply is added to the appropriatecommunity
Border filters already in place take care of announcements⇒ Ease of operation!
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 118
Community Example: Internet Edge
This demonstrates how communities might be used atthe peering edge of an ISP network
ISP has four types of BGP peers:CustomerIXP peerPrivate peerTransit provider
The prefixes received from each can be classified usingcommunities
Customers can opt to receive any or all of the above
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 119
Community Example: Internet Edge
Community assignments:Customer prefix: community 100:3000IXP prefix: community 100:3100Private peer prefix: community 100:3200
BGP customer who buys local connectivity gets 100:3000 BGP customer who buys local and IXP connectivity receives
community 100:3000 and 100:3100 BGP customer who buys full peer connectivity receives community
100:3000, 100:3100, and 100:3200 Customer who wants “the Internet” gets everything
Gets default route originated by aggregation routerOr pays money to get all 220k prefixes
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 120
Community Example: Internet Edge
No need to create customised filters when addingcustomers
Border router already sets communitiesInstallation engineers pick the appropriate community set whenestablishing the customer BGP session⇒ Ease of operation!
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 121
Community Example – Summary
Two examples of customer edge and internet edge canbe combined to form a simple community solution forISP prefix policy control
More experienced operators tend to have moresophisticated options available
Advice is to start with the easy examples given, and thenproceed onwards as experience is gained
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 122
ISP BGP Communities
There are no recommended ISP BGP communities apart fromRFC1998The five standard communities
www.iana.org/assignments/bgp-well-known-communities
Efforts have been made to document from time to timetotem.info.ucl.ac.be/publications/papers-elec-versions/draft-quoitin-bgp-comm-survey-00.pdfBut so far… nothing more… Collection of ISP communities at www.onesc.net/communitiesNANOG Tutorial:www.nanog.org/meetings/nanog40/presentations/BGPcommunities.pdf
ISP policy is usually publishedOn the ISP’s websiteReferenced in the AS Object in the IRR
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 123
ISP Examples: Sprint
More info athttps://www.sprint.net/index.php?p=policy_bgp
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 124
Some ISP Examples:NTT
More info atwww.us.ntt.net/about/policy/routing.cfm
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 125
ISP Examples:Verizon Business Europe
aut-num: AS702descr: Verizon Business EMEA - Commercial IP service provider in Eurremarks: VzBi uses the following communities with its customers: 702:80 Set Local Pref 80 within AS702 702:120 Set Local Pref 120 within AS702 702:20 Announce only to VzBi AS'es and VzBi customers 702:30 Keep within Europe, don't announce to other VzBi AS 702:1 Prepend AS702 once at edges of VzBi to Peers 702:2 Prepend AS702 twice at edges of VzBi to Peers 702:3 Prepend AS702 thrice at edges of VzBi to Peers Advanced communities for customers 702:7020 Do not announce to AS702 peers with a scope of National but advertise to Global Peers, European Peers and VzBi customers. 702:7001 Prepend AS702 once at edges of VzBi to AS702 peers with a scope of National. 702:7002 Prepend AS702 twice at edges of VzBi to AS702 peers with a scope of National.(more)
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 126
ISP Examples:Verizon Business Europe
(more) 702:7003 Prepend AS702 thrice at edges of VzBi to AS702 peers with a scope of National. 702:8020 Do not announce to AS702 peers with a scope of European but advertise to Global Peers, National Peers and VzBi customers. 702:8001 Prepend AS702 once at edges of VzBi to AS702 peers with a scope of European. 702:8002 Prepend AS702 twice at edges of VzBi to AS702 peers with a scope of European. 702:8003 Prepend AS702 thrice at edges of VzBi to AS702 peers with a scope of European. -------------------------------------------------------------- Additional details of the VzBi communities are located at: http://www.verizonbusiness.com/uk/customer/bgp/ --------------------------------------------------------------mnt-by: WCOM-EMEA-RICE-MNTsource: RIPE
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 127
Some ISP ExamplesBT Ignite
aut-num: AS5400descr: BT Ignite European Backboneremarks:remarks: Community to Community toremarks: Not announce To peer: AS prepend 5400remarks:remarks: 5400:1000 All peers & Transits 5400:2000remarks:remarks: 5400:1500 All Transits 5400:2500remarks: 5400:1501 Sprint Transit (AS1239) 5400:2501remarks: 5400:1502 SAVVIS Transit (AS3561) 5400:2502remarks: 5400:1503 Level 3 Transit (AS3356) 5400:2503remarks: 5400:1504 AT&T Transit (AS7018) 5400:2504remarks: 5400:1506 GlobalCrossing Trans(AS3549) 5400:2506remarks:remarks: 5400:1001 Nexica (AS24592) 5400:2001remarks: 5400:1002 Fujitsu (AS3324) 5400:2002remarks: 5400:1004 C&W EU (1273) 5400:2004<snip>notify: [email protected]: CIP-MNTsource: RIPE
And manymany more!
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 128
Some ISP ExamplesLevel 3
aut-num: AS3356descr: Level 3 Communications<snip>remarks: -------------------------------------------------------remarks: customer traffic engineering communities - Suppressionremarks: -------------------------------------------------------remarks: 64960:XXX - announce to AS XXX if 65000:0remarks: 65000:0 - announce to customers but not to peersremarks: 65000:XXX - do not announce at peerings to AS XXXremarks: -------------------------------------------------------remarks: customer traffic engineering communities - Prependingremarks: -------------------------------------------------------remarks: 65001:0 - prepend once to all peersremarks: 65001:XXX - prepend once at peerings to AS XXX<snip>remarks: 3356:70 - set local preference to 70remarks: 3356:80 - set local preference to 80remarks: 3356:90 - set local preference to 90remarks: 3356:9999 - blackhole (discard) traffic<snip>mnt-by: LEVEL3-MNTsource: RIPE And many
many more!
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 129
BGP for Internet Service Providers
BGP Basics
Scaling BGP
Using Communities
Deploying BGP in an ISP network
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 130
Deploying BGP in an ISP Network
Okay, so we’ve learned all about BGP now; how do weuse it on our network??
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 131
Deploying BGP
The role of IGPs and iBGP
Aggregation
Receiving Prefixes
Configuration Tips
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 132
The role of IGP and iBGP
Ships in the night?OrGood foundations?
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 133
BGP versus OSPF/ISIS
Internal Routing Protocols (IGPs)examples are ISIS and OSPFused for carrying infrastructure addressesNOT used for carrying Internet prefixes or customer prefixesdesign goal is to minimise number of prefixes in IGP to aidscalability and rapid convergence
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 134
BGP versus OSPF/ISIS
BGP used internally (iBGP) and externally (eBGP) iBGP used to carry
some/all Internet prefixes across backbonecustomer prefixes
eBGP used toexchange prefixes with other ASesimplement routing policy
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 135
BGP/IGP model used in ISP networks
Model representation
IGP
iBGP
IGP
iBGP
IGP
iBGP
IGP
iBGP
eBGP eBGP eBGP
AS1 AS2 AS3 AS4
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 136
BGP versus OSPF/ISIS
DO NOT:distribute BGP prefixes into an IGPdistribute IGP routes into BGPuse an IGP to carry customer prefixes
YOUR NETWORK WILL NOT SCALE
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 137
Injecting prefixes into iBGP
Use iBGP to carry customer prefixesDon’t ever use IGP
Point static route to customer interface Enter network into BGP process
Ensure that implementation options are used so that the prefixalways remains in iBGP, regardless of state of interfacei.e. avoid iBGP flaps caused by interface flaps
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 139
Aggregation
Aggregation means announcing the address blockreceived from the RIR to the other ASes connected toyour network
Subprefixes of this aggregate may be:Used internally in the ISP networkAnnounced to other ASes to aid with multihoming
Unfortunately too many people are still thinking aboutclass Cs, resulting in a proliferation of /24s in theInternet routing table
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 140
Aggregation
Address block should be announced to the Internet asan aggregate
Subprefixes of address block should NOT beannounced to Internet unless for traffic engineeringpurposes
(see BGP Multihoming Tutorial)
Aggregate should be generated internallyNot on the network borders!
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 141
Announcing an Aggregate
ISPs who don’t and won’t aggregate are held in poorregard by community
Registries publish their minimum allocation sizeAnything from a /20 to a /22 depending on RIRDifferent sizes for different address blocks
No real reason to see anything longer than a /22 prefixin the Internet
BUT there are currently >180000 /24s!
But: APNIC changed (Oct 2010) its minimum allocationsize on all blocks to /24
IPv4 run-out is starting to have an impact
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 142
AS100customer
100.10.10.0/23Internet
100.10.10.0/23100.10.0.0/24100.10.4.0/22…
Aggregation – Example
Customer has /23 network assigned from AS100’s /19 address block
AS100 announces customers’ individual networks to the Internet
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 143
Customer link returnsTheir /23 network is nowvisible to their ISPTheir /23 network is re-advertised to peersStarts rippling through InternetLoad on Internet backbonerouters as network isreinserted into routing tableSome ISP’s suppress the flapsInternet may take 10-20 min orlonger to be visibleWhere is the Quality ofService???
Customer link goes downTheir /23 network becomesunreachable/23 is withdrawn from AS100’siBGP
Their ISP doesn’t aggregate its/19 network block
/23 network withdrawalannounced to peersstarts rippling through theInternetadded load on all Internetbackbone routers as networkis removed from routing table
Aggregation – Bad Example
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 144
AS100customer
100.10.10.0/23
100.10.0.0/19aggregate
Internet
100.10.0.0/19
Aggregation – Example
Customer has /23 network assigned from AS100’s /19 address block
AS100 announced /19 aggregate to the Internet
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 145
Aggregation – Good Example
Customer link goes downtheir /23 network becomesunreachable/23 is withdrawn from AS100’siBGP
/19 aggregate is still beingannounced
no BGP hold down problemsno BGP propagation delaysno damping by other ISPs
Customer link returns
Their /23 network is visibleagain
The /23 is re-injected intoAS100’s iBGP
The whole Internet becomesvisible immediately
Customer has Quality ofService perception
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 146
Aggregation – Summary
Good example is what everyone should do!Adds to Internet stabilityReduces size of routing tableReduces routing churnImproves Internet QoS for everyone
Bad example is what too many still do!Why? Lack of knowledge?Laziness?
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 147
Separation of iBGP and eBGP
Many ISPs do not understand the importance ofseparating iBGP and eBGP
iBGP is where all customer prefixes are carriedeBGP is used for announcing aggregate to Internet and forTraffic Engineering
Do NOT do traffic engineering with customer originatediBGP prefixes
Leads to instability similar to that mentioned in the earlier badexampleEven though aggregate is announced, a flapping subprefix willlead to instability for the customer concerned
Generate traffic engineering prefixes on the BorderRouter
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 148
The Internet Today (February 2011)
Current Internet Routing Table StatisticsBGP Routing Table Entries 345357Prefixes after maximum aggregation 155769Unique prefixes in Internet 170883Prefixes smaller than registry alloc 142680/24s announced 180715ASes in use 35825
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 149
“The New Swamp”
Swamp space is name used for areas of pooraggregation
The original swamp was 192.0.0.0/8 from the former class Cblock
Name given just after the deployment of CIDR
The new swamp is creeping across all parts of the InternetNot just RIR space, but “legacy” space too
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 150
“The New Swamp”RIR Space – February 1999
RIR blocks contribute 88% of the Internet Routing TableBlock Networks118/8 0119/8 0120/8 0121/8 0122/8 0123/8 0124/8 0125/8 0126/8 0173/8 0174/8 0186/8 0187/8 0189/8 0190/8 0192/8 6275193/8 2390194/8 2932195/8 1338196/8 513198/8 4034199/8 3495200/8 1348
Block Networks201/8 0202/8 2276203/8 3622204/8 3792205/8 2584206/8 3127207/8 2723208/8 2817209/8 2574210/8 617211/8 0212/8 717213/8 1216/8 943217/8 0218/8 0219/8 0220/8 0221/8 0222/8 0
Block Networks24/8 16541/8 058/8 059/8 060/8 061/8 362/8 8763/8 2064/8 065/8 066/8 067/8 068/8 069/8 070/8 071/8 072/8 073/8 074/8 075/8 076/8 077/8 078/8 0
Block Networks79/8 080/8 081/8 082/8 083/8 084/8 085/8 086/8 087/8 088/8 089/8 090/8 091/8 096/8 097/8 098/8 099/8 0112/8 0113/8 0114/8 0115/8 0116/8 0117/8 0
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 151
“The New Swamp”RIR Space – February 2010
Block Networks118/8 1349119/8 1694120/8 531121/8 1756122/8 2687123/8 2400124/8 2259125/8 2514126/8 106173/8 1994174/8 1089186/8 1223187/8 1501189/8 3063190/8 6945192/8 6952193/8 6820194/8 5177195/8 5325196/8 1857198/8 4504199/8 4372 200/8 8884
Block Networks201/8 4136202/8 11354203/8 11677204/8 5744205/8 3037206/8 3951207/8 4635208/8 6498209/8 5536210/8 4977211/8 3130212/8 3550213/8 3442216/8 7645217/8 3136218/8 1512219/8 1303220/8 2108221/8 980222/8 1058
Block Networks24/8 332841/8 344858/8 167559/8 157560/8 88861/8 289062/8 241863/8 311464/8 660165/8 396666/8 778267/8 377168/8 322169/8 528070/8 200871/8 132772/8 405073/8 474/8 507475/8 116476/8 103477/8 196478/8 1397
Block Networks79/8 111980/8 233581/8 170982/8 135883/8 135784/8 134185/8 249286/8 78087/8 146688/8 106889/8 316890/8 37791/8 455596/8 77897/8 72598/8 131299/8 288112/8 883113/8 890114/8 996115/8 1616116/8 1755117/8 1611
RIR blocks contribute about 87% of the Internet Routing Table
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 152
“The New Swamp”Summary
RIR space shows creeping deaggregationIt seems that an RIR /8 block averages around 5000 prefixes(and upwards) once fully allocated
Food for thought:The 120 RIR /8s combined will cause:635000 prefixes with 5000 prefixes per /8 density762000 prefixes with 6000 prefixes per /8 densityPlus 12% due to “non RIR space deaggregation”→ Routing Table size of 853440 prefixes
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 153
“The New Swamp”Summary
Rest of address space is showing similar deaggregationtoo
What are the reasons?Main justification is traffic engineering
Real reasons are:Lack of knowledgeLazinessDeliberate & knowing actions
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 154
Efforts to improve aggregation
The CIDR ReportInitiated and operated for many years by Tony BatesNow combined with Geoff Huston’s routing analysis
www.cidr-report.orgResults e-mailed on a weekly basis to most operations listsaround the worldLists the top 30 service providers who could do better ataggregating
RIPE Routing WG aggregation recommendationRIPE-399 — http://www.ripe.net/ripe/docs/ripe-399.html
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 155
Efforts to Improve AggregationThe CIDR Report
Also computes the size of the routing table assumingISPs performed optimal aggregation
Website allows searches and computations ofaggregation to be made on a per AS basis
Flexible and powerful tool to aid ISPsIntended to show how greater efficiency in terms of BGP tablesize can be obtained without loss of routing and policyinformationShows what forms of origin AS aggregation could be performedand the potential benefit of such actions to the total table sizeVery effectively challenges the traffic engineering excuse
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 162
Importance of Aggregation
Size of routing tableRouter Memory is not so much of a problem as it was in the1990sRouters can be specified to carry 1 million+ prefixes
Convergence of the Routing SystemThis is a problemBigger table takes longer for CPU to processBGP updates take longer to deal withBGP Instability Report tracks routing system update activityhttp://bgpupdates.potaroo.net/instability/bgpupd.html
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 165
Aggregation Potential(source: bgp.potaroo.net/as2.0/)
AS Path
AS Origin
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 166
AggregationSummary
Aggregation on the Internet could be MUCH better35% saving on Internet routing table size is quite feasibleTools are available
Commands on the routers are not hardCIDR-Report webpage
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 168
Receiving Prefixes
There are three scenarios for receiving prefixes fromother ASNs
Customer talking BGPPeer talking BGPUpstream/Transit talking BGP
Each has different filtering requirements and need to beconsidered separately
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 169
Receiving Prefixes:From Customers
ISPs should only accept prefixes which have beenassigned or allocated to their downstream customer
If ISP has assigned address space to its customer, thenthe customer IS entitled to announce it back to his ISP
If the ISP has NOT assigned address space to itscustomer, then:
Check the five RIR databases to see if this address space reallyhas been assigned to the customerThe tool: whois
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 170
Example use of whois to check if customer is entitled to announceaddress space:
$ whois -h whois.apnic.net 202.12.29.0inetnum: 202.12.28.0 - 202.12.29.255netname: APNIC-APdescr: Asia Pacific Network Information Centerdescr: Level 1 - 33 Park Road.descr: Milton QLD 4064descr: Australiacountry: AUadmin-c: AIC1-APtech-c: NO4-APmnt-by: APNIC-HMchanged: [email protected] 19980918status: ASSIGNED PORTABLEsource: APNIC
Portable – means its an assignmentto the customer, the customer canannounce it to you
Receiving Prefixes:From Customers
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 171
Example use of whois to check if customer is entitled to announce addressspace:
$ whois -h whois.ripe.net 193.128.0.0inetnum: 193.128.0.0 - 193.133.255.255netname: UK-PIPEX-193-128-133descr: Verizon UK Limitedcountry: GBorg: ORG-UA24-RIPEadmin-c: WERT1-RIPEtech-c: UPHM1-RIPEstatus: ALLOCATED UNSPECIFIEDremarks: Please send abuse notification to [email protected]: RIPE-NCC-HM-MNTmnt-lower: AS1849-MNTmnt-routes: AS1849-MNTmnt-routes: WCOM-EMEA-RICE-MNTmnt-irt: IRT-MCI-GBsource: RIPE # Filtered
ALLOCATED – means that this isProvider Aggregatable address spaceand can only be announced by the ISPholding the allocation (in this caseVerizon UK)
Receiving Prefixes:From Customers
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 172
Receiving Prefixes:From Peers
A peer is an ISP with whom you agree to exchangeprefixes you originate into the Internet routing table
Prefixes you accept from a peer are only those they haveindicated they will announcePrefixes you announce to your peer are only those you haveindicated you will announce
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 173
Receiving Prefixes:From Peers
Agreeing what each will announce to the other:Exchange of e-mail documentation as part of the peeringagreement, and then ongoing updates
ORUse of the Internet Routing Registry and configuration toolssuch as the IRRToolSet
www.isc.org/sw/IRRToolSet/
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 174
Receiving Prefixes:From Upstream/Transit Provider
Upstream/Transit Provider is an ISP who you pay togive you transit to the WHOLE Internet
Receiving prefixes from them is not desirable unlessreally necessary
Traffic Engineering – see BGP Multihoming Tutorial
Ask upstream/transit provider to either:originate a default-route
ORannounce one prefix you can use as default
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 175
Receiving Prefixes:From Upstream/Transit Provider
If necessary to receive prefixes from any provider, careis required.
Don’t accept default (unless you need it)Don’t accept your own prefixes
For IPv4:Don’t accept private (RFC1918) and certain special useprefixes:
http://www.rfc-editor.org/rfc/rfc5735.txtDon’t accept prefixes longer than /24 (?)
For IPv6:Don’t accept certain special use prefixes:
http://www.rfc-editor.org/rfc/rfc5156.txtDon’t accept prefixes longer than /48 (?)
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 176
Receiving Prefixes:From Upstream/Transit Provider
Check Team Cymru’s list of “bogons”www.team-cymru.org/Services/Bogons/http.html
For IPv4 also consult:www.ietf.org/internet-drafts/draft-vegoda-no-more-unallocated-slash8s-00.txt
For IPv6 also consult:www.space.net/~gert/RIPE/ipv6-filters.html
Bogon Route Server:www.team-cymru.org/Services/Bogons/routeserver.htmlSupplies a BGP feed (IPv4 and/or IPv6) of address blockswhich should not appear in the BGP table
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 177
Receiving Prefixes
Paying attention to prefixes received from customers,peers and transit providers assists with:
The integrity of the local networkThe integrity of the Internet
Responsibility of all ISPs to be good Internet citizens
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 178
Configuration Tips
Of passwords, tricks and templates
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 179
iBGP and IGPsReminder!
Make sure loopback is configured on routeriBGP between loopbacks, NOT real interfaces
Make sure IGP carries loopback /32 address
Consider the DMZ nets:Use unnumbered interfaces?Use next-hop-self on iBGP neighboursOr carry the DMZ /30s in the iBGPBasically keep the DMZ nets out of the IGP!
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 180
iBGP: Next-hop-self
BGP speaker announces external network to iBGPpeers using router’s local address (loopback) as next-hop
Used by many ISPs on edge routersPreferable to carrying DMZ /30 addresses in the IGPReduces size of IGP to just core infrastructureAlternative to using unnumbered interfacesHelps scale networkMany ISPs consider this “best practice”
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 181
Limiting AS Path Length
Some BGP implementations have problems with longAS_PATHS
Memory corruptionMemory fragmentation
Even using AS_PATH prepends, it is not normal to seemore than 20 ASes in a typical AS_PATH in theInternet today
The Internet is around 5 ASes deep on averageLargest AS_PATH is usually 16-20 ASNs
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 182
Limiting AS Path Length
Some announcements have ridiculous lengths of AS-paths:
*> 3FFE:1600::/24 22 11537 145 12199 1031810566 13193 1930 2200 3425 293 5609 5430 13285 693914277 1849 33 15589 25336 6830 8002 2042 7610 i
This example is an error in one IPv6 implementation
*> 96.27.246.0/24 2497 1239 12026 12026 1202612026 12026 12026 12026 12026 12026 12026 1202612026 12026 12026 12026 12026 12026 12026 1202612026 12026 12026 i
This example shows 21 prepends (for no obvious reason)
If your implementation supports it, consider limiting themaximum AS-path length you will accept
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 183
BGP TTL “hack”
Implement RFC5082 on BGP peerings(Generalised TTL Security Mechanism)Neighbour sets TTL to 255Local router expects TTL of incoming BGP packets to be 254No one apart from directly attached devices can send BGPpackets which arrive with TTL of 254, so any possible attack bya remote miscreant is dropped due to TTL mismatch
ISP AS 100Attacker
TTL 254
TTL 253 TTL 254R1 R2
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 184
BGP TTL “hack”
TTL Hack:Both neighbours must agree to use the featureTTL check is much easier to perform than MD5(Called BTSH – BGP TTL Security Hack)
Provides “security” for BGP sessionsIn addition to packet filters of courseMD5 should still be used for messages which slip through theTTL hackSee www.nanog.org/mtg-0302/hack.html for more details
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 185
Templates
Good practice to configure templates for everythingVendor defaults tend not to be optimal or even very useful forISPsISPs create their own defaults by using configuration templates
eBGP and iBGP examples followAlso see Team Cymru’s BGP templates
http://www.team-cymru.org/ReadingRoom/Documents/
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 186
iBGP TemplateExample
iBGP between loopbacks!
Next-hop-selfKeep DMZ and external point-to-point out of IGP
Always send communities in iBGPOtherwise accidents will happen
Hardwire BGP to version 4Yes, this is being paranoid!
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 187
iBGP TemplateExample continued
Use passwords on iBGP sessionNot being paranoid, VERY necessaryIt’s a secret shared between you and your peerIf arriving packets don’t have the correct MD5 hash, they areignoredHelps defeat miscreants who wish to attack BGP sessions
Powerful preventative tool, especially when combinedwith filters and the TTL “hack”
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 188
eBGP TemplateExample
BGP dampingDo NOT use it unless you understand the impactDo NOT use the vendor defaults without thinking
Remove private ASes from announcementsCommon omission today
Use extensive filters, with “backup”Use as-path filters to backup prefix filtersKeep policy language for implementing policy, rather than basicfiltering
Use password agreed between you and peer on eBGPsession
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 189
eBGP TemplateExample continued
Use maximum-prefix trackingRouter will warn you if there are sudden increases in BGP tablesize, bringing down eBGP if desired
Limit maximum as-path length inbound
Log changes of neighbour state…and monitor those logs!
Make BGP admin distance higher than that of any IGPOtherwise prefixes heard from outside your network couldoverride your IGP!!
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 190
Summary
Use configuration templates
Standardise the configuration
Be aware of standard “tricks” to avoid compromise ofthe BGP session
Anything to make your life easier, network less prone toerrors, network more likely to scale
It’s all about scaling – if your network won’t scale, thenit won’t be successful
© 2011 Cisco Systems, Inc. All rights reserved.APRICOT 2011 191
BGP Techniques for Internet ServiceProviders
Philip Smith <[email protected]>APRICOT 2011Hong Kong, SAR, China15 - 25 February 2011